Building an AI-assisted system for creating, explaining, and fixing formulas

DEVELOPER TOOLS | ENTERPRISE B2B SAAS 

 

PROJECT OVERVIEW

OVERVIEW

Formulas are powerful but notoriously hard to create, debug, and understand. This Unqork project explored how AI could act as a collaborative assistant inside a technical creation workflow, helping users build and fix formulas using natural language while preserving control and trust.

THE RESULT

an AI Formula Companion: a contextual popover that lives directly inside a formula cell and supports creation, explanation, and error recovery without forcing users to leave their work.

SKILLS

AI Interaction Design
System & State Modeling
UX for Technical Workflows
Error Handling & Guardrails

ROLE

Lead Product Designer

TIMELINE

3 Weeks (September 2023)

PROBLEM

Users building formulas face three persistent challenges:

  1. High cognitive load: Complex syntax and nested logic are difficult to reason about.

  2. Error-prone workflows: Small mistakes lead to broken formulas and frustrating debugging cycles.

  3. Fragmented help: Users jump between product docs, forums, and external references to solve a single issue.

Insight: Formulas represent structured intent, making them an ideal candidate for AI-assisted translation from natural language to logic.

DESIGN RESEARCH

This project built on a prior Formula Autocomplete initiative I led, which gave me existing research on how creators approach formula building across Excel, Tableau, and Unqork. I supplemented this with competitive analysis of AI tools (Google Duet, Notion AI) and formula builders (Google Sheets, Coda).

Insight: the most effective AI tools are contextual and assistive, not separate chat destinations. Users don't want to leave their workflow to get help — they want help to come to them.

Previous autocomplete artifacts:

DESIGN PROCESS

Building a Reliable Human<>AI Workflow

I mapped five core states the AI needed to support: opening the builder, creating a formula from natural language, explaining an existing formula, safely modifying a working one, and diagnosing a broken one.

Because Unqork had no existing chatbot UI, I created new interaction patterns from scratch — recommended prompts, user inputs, and AI responses — each visually distinct to reinforce clarity and trust.

Working closely with platform engineers, I ensured AI suggestions were non-destructive and reversible. The design philosophy shifted from novelty to predictability, which is critical for AI in technical workflows.

Flow 1: Create a formula using natural language

Flow 2: Get help changing an existing formula

Flow 3: Fix a broken formula in seconds

OUTCOME

The proof of concept shipped and was showcased at AWS re:Invent 2023, demonstrating a working AI-assisted formula experience and a scalable interaction model for AI-native creation tools.

Previous
Previous

Increasing demo-to-close conversion rates by streamlining decision-making

Next
Next

Envisioning the Netflix of Personal Finance